Dose-time-concentration prediction method based on GRU-TCN with temporal-channel attention

Area of Science:

• Pharmacokinetics and Pharmacodynamics
• Machine Learning in Drug Development
• Computational Biology

Background:

• Accurate dose-time-concentration prediction is crucial for precision dosing.
• Existing models struggle with data scarcity and multicollinearity in pharmacokinetic data.
• Augmenting limited data while preserving distribution consistency is a significant challenge.

Purpose of the Study:

• To develop a robust model for dose-time-concentration prediction under challenging data conditions.
• To introduce a novel GRU-TCN model with Temporal-Channel Attention (GT-TCA) for enhanced prediction accuracy.
• To validate the model's performance on both real-world (Buyang Huanwu Decoction) and simulated pharmacokinetic datasets.

Main Methods:

• Utilized a Gated Recurrent Unit-Temporal Convolutional Network (GRU-TCN) architecture.
• Incorporated Temporal-Channel Attention (GT-TCA) to focus on informative time steps and analytes.
• Employed TimeCVAE for augmenting limited pharmacokinetic data with distribution-consistent sequences.
• Evaluated performance using Mean Absolute Error (MAE) and R-squared (R²) metrics.

Main Results:

• The GT-TCA model achieved a 22.7% reduction in MAE and a 4% improvement in R² compared to baseline models.
• Ablation studies confirmed that the attention mechanism reduced MAE and RMSE by 6% and 5%, respectively.
• The model demonstrated superior performance on Buyang Huanwu Decoction (normal/inflammatory) and simulation datasets (RG1678, RIF).

Conclusions:

• The proposed GT-TCA model effectively addresses data scarcity and multicollinearity in dose-time-concentration prediction.
• The model provides more precise quantitative evidence to support precision dosing strategies.
• The integration of GRU, TCN, and attention mechanisms offers a robust solution for pharmacokinetic data analysis.